基于窗口滤波与均值滤波的深度图像实时修复算法

A Depth Image Real-Time Image Inpainting Algorithm Based on Window Filtering and Mean Filtering

使用Kinect作为视频图像的输入设备,获取的连续帧的深度图像会在边缘出现空洞像素点和像素点随机抖动的问题。本文针对Kinect获取的连续帧深度图像提出了加权窗口-除零均值滤波算法。先使用提出的加权窗口滤波算法对空洞像素填补,然后采用改进后的除零均值滤波算法对图像进行平滑与去噪声处理。实验结果表明,相比双边滤波、中值滤波、均值滤波和高斯滤波算法,本文提出的算法能够有效的减少图像中的抖动像素,去除深度图像中的噪声信息,使图像边缘信息平滑,并且保证视频输入有较高的帧率。

As an input device of video images,Kinect can obtain the depth images of successive frames.However,there exists the issue of edge holes and random jitter of pixels in the depth video images.In this paper,the weighted window flter and the non-zero mean fltering algorithm are proposed for the successive frame depth image acquired by Kinect.Firstly,the depth video image is repaired by the proposed weighted window fltering algorithm.The image is then smoothed and the noise is eliminated by using an improved non-zero mean fltering algorithm.The experimental results show that the algorithm can more effectively remove a large number of jitter pixels in video images and smooth edge information than the algorithms of bilateral fltering,median fltering,mean fltering and Gaussian fltering.The algorithm ensures a high frame rate of the video input.